Exploring early fetal brain development: a deep learning approach

博士项目目标：在这个项目中，我们将开发深度学习图像分析工具，以了解多模态MRI所描述的早期大脑发育。我们将开发方法来描述瞬态和新兴的大脑结构，并在平均健康的大脑发育和个体水平上表征它们的微观结构。我们将提取正常和异常早期大脑发育的相关生物标志物，包括已知与癫痫和自闭症谱系障碍有关的解剖结构。项目描述/背景：在怀孕的后半期，大脑经历快速发育，包括白质束的形成，髓鞘形成和皮质折叠的开始（图1）。了解这些发育过程的精确时间和变化将有助于了解各种疾病的起源，如自闭症谱系障碍、癫痫，以及子宫内先天性异常或感染的影响。使用多模态MRI进行神经成像可以深入了解胎儿大脑发育[1,2]。然而，对母亲体内运动的胎儿进行成像是困难的，这导致图像质量不稳定，而且大脑的快速发育会干扰胎儿脑MRI图像的解释，这需要专门的胎儿图像分析技术[3]。在这个项目中，我们建议建立深度学习工具，专门用于分析多模态MRI观察到的胎儿早期大脑快速发育。我们将建立在发展人类连接组项目上，该项目产生了一个大型的最先进的运动校正多模态胎儿MRI数据库。我们将使用人工智能来检测和描绘快速进化的瞬态胎儿结构，通过融合从结构和扩散MRI中提取的形态学，微观结构和连接信息。开发先进的深度学习技术，用于胎儿MRI图像增强，以减少伪影，并能够对个体婴儿进行可靠的定量评估。开发海马（旋转/折叠指数）和Sylvian裂缝（包皮化指数）等大脑区域结构发育的定量指标。开发时空深度学习模型来预测ASD和早产的风险，并对这些模型进行解释，以找到表征这些疾病的多模态MRI生物标志物。

Aim of the PhD Project:In this project we will develop deep learning image analysis tools to enable understanding of early brain development as depicted by multi-modal MRI. We will develop methods to delineate transient and emerging brain structures and characterise their microstructure in average healthy brain development as well as on individual level. We will extract relevant biomarkers of normal and abnormal early brain development, including anatomy known to be involved in epilepsy and autism spectrum disorder. Project Description / Background:During the second half of pregnancy the brain undergoes rapid development, including formation of white matter tracts, the onset of myelination and cortical folding (Figure 1). Understanding precise timing and variation of these developmental processes would shed light on origins of various conditions, such as Autism Spectrum Disorder, Epilepsy, and effects of congenital abnormalities or infection in in womb. Neuroimaging using multi-modal MRI can offer insights into fetal brain development [1,2]. However, imaging moving fetus inside the mother is difficult, resulting in variable image quality, and fast brain development interferes with interpretation of the fetal brain MRI images, requiring dedicated fetal image analysis techniques [3]. In this project we propose to build deep learning tools dedicated to analysis of rapid early fetal brain development as observed by multi-modal MRI. We will build on Developing Human Connectome project that produced a large database of state-of-the-art motion-corrected multi-modal fetal MRI [4]. We will use artificial intelligence to Detect and delineate rapidly evolving transient fetal structures, by fusing morphological, microstructural and connectivity information extracted from structural and diffusion MRI. Derive quantitative indices, including volumes, shape and microstructure to accurately stage the fetal brain development Develop advance deep learning techniques for fetal MRI image enhancement to reduce artefacts and enable reliable quantitative assessment of individual babies Derive quantitative indices of structural development for brain regions like hippocampus (rotation/folding indices) and Sylvian fissure (opercularisation indices), known to be involved in Epilepsy Develop spatio-temporal deep learning models to predict risk of ASD and preterm birth and interpret these models to find multi-modal MRI biomarkers that characterise these conditions.